Received-SPF: pass (google.com: domain of linux-kernel+bounces-141466-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:45e3:2400::1 as permitted sender) client-ip=2604:1380:45e3:2400::1;
Date: Thu, 11 Apr 2024 20:17:07 +0200 (CEST)
From: Julia Lawall <julia.lawall@inria.fr>
To: Greg KH <gregkh@linuxfoundation.org>
cc: Roman Storozhenko <romeusmeister@gmail.com>, jirislaby@kernel.org, 
    skhan@linuxfoundation.org, javier.carrasco.cruz@gmail.com, 
    linux-kernel@vger.kernel.org, linux-serial@vger.kernel.org
Subject: Re: [PATCH] sysrq: Auto release device node using __free attribute
In-Reply-To: <2024041111-tummy-boil-a6aa@gregkh>
Message-ID: <b6b33aa-faef-6919-7125-c2db11e784ee@inria.fr>
References: <20240411180256.61001-1-romeusmeister@gmail.com> <2024041111-tummy-boil-a6aa@gregkh>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII


On Thu, 11 Apr 2024, Greg KH wrote:

> On Thu, Apr 11, 2024 at 08:02:56PM +0200, Roman Storozhenko wrote:
> > Add a cleanup function attribute '__free(device_node)' to the device node
> > pointer initialization statement and remove the pairing cleanup function
> > call of 'of_node_put' at the end of the function.
> > The '_free()' attrubute is introduced by scope-based resource management
> > in-kernel framework implemented in 'cleanup.h'. A pointer marked with
> > '__free()' attribute makes a compiler insert a cleanup function call
> > to the places where the pointer goes out of the scope. This feature
> > allows to get rid of manual cleanup function calls.
> >
> > Suggested-by: Julia.Lawall <Julia.Lawall@inria.fr>
> > Signed-off-by: Roman Storozhenko <romeusmeister@gmail.com>
> > ---
> > This patch targets the next tree:
> > tree: git://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git
> > tag: next-20240411
> > ---
> >  drivers/tty/sysrq.c | 7 +++----
> >  1 file changed, 3 insertions(+), 4 deletions(-)
> >
> > diff --git a/drivers/tty/sysrq.c b/drivers/tty/sysrq.c
> > index 02217e3c916b..1d1261f618c0 100644
> > --- a/drivers/tty/sysrq.c
> > +++ b/drivers/tty/sysrq.c
> > @@ -758,11 +758,12 @@ static void sysrq_detect_reset_sequence(struct sysrq_state *state,
> >  static void sysrq_of_get_keyreset_config(void)
> >  {
> >  	u32 key;
> > -	struct device_node *np;
> >  	struct property *prop;
> >  	const __be32 *p;
> >
> > -	np = of_find_node_by_path("/chosen/linux,sysrq-reset-seq");
> > +	struct device_node *np __free(device_node) =
> > +		of_find_node_by_path("/chosen/linux,sysrq-reset-seq");
> > +
> >  	if (!np) {
> >  		pr_debug("No sysrq node found");
> >  		return;
> > @@ -781,8 +782,6 @@ static void sysrq_of_get_keyreset_config(void)
> >
> >  	/* Get reset timeout if any. */
> >  	of_property_read_u32(np, "timeout-ms", &sysrq_reset_downtime_ms);
> > -
> > -	of_node_put(np);
> >  }
> >  #else
> >  static void sysrq_of_get_keyreset_config(void)
>
> Also, this change really makes no sense at all, the pointer never goes
> out of scope except when the function is over, at the bottom.  So why
> make this complex change at all for no benefit?
>
> In other words, properly understand the change you are making and only
> make it if it actually makes sense.  It does not make any sense here,
> right?

Maybe it would be nice to get rid of of_node_puts in the general case?
Even though this one is not very annoying, there are some other functions
where there are many of_node_puts, and convoluted error handling code to
incorporate them on both the success and failure path.  So maybe it would
be better to avoid the situation of having them sometimes and not having
them other times?  But this is an opinion, and if the general consensus is
to only get rid of the cases that currently add complexity, then that is
possible too.

julia